Discovery
Boundary Case Atlas
36 counterexamples recorded before evaluation, by project. A method that passes its headline metric and fails one of these has not met the project’s question.
The boundary cases are design requirements. The synthetic checks beside them test a subset in code; the rest wait for eligible data.
Discovery and data
Structure-Based AI Small-Molecule Discovery
- Evaluation leakage across molecular scaffolds and homologous target families
- Inconsistent assay conditions and units across binding measurements
- The research specification also sets limits of 0.4 structural similarity, 2.0 Å RMSD over 100 ns and five retrosynthetic steps
Early Boost for Core Health R&D Technology
- Original rows or repeated records for the same person appearing across splits
- Differential privacy budget and proximity of synthetic rows to source records
- An internal target of zero biologically impossible variable combinations
Rare Disease Diagnostic Research
- Gene ranks in synthetic cases containing known variants
- Ranking and evidence changes after missing or miscoded phenotypes
- Reference build mismatches, conflicting classifications and missing literature references
Cognitive and neurological health
Non-Pharmacological Dementia Research
- Convergence to 70–80% correct responses in simulated episodes
- Difficulty changes after repeated errors, fatigue or delayed responses
- 100% blocking of phrases tagged for negative emotional effects in the fixed test
Multimodal Dementia Diagnostic Research
- Whether MRI and speech can be matched by participant and visit
- Performance changes across scanners, sites and audio settings
- Lock model weights and diagnostic definitions before external evaluation
Patient safety
Medication Administration and Dispensing Safety
- Look-alike products under glare, distorted lighting and overlap
- Alerts for sound-alike products and unit mismatches
- Abstention when required lab values are missing or conflicting
Fall Prediction and Prevention
- Event-level assessment of a warning at least three seconds before impact
- False alarms per hour from movements such as picking up objects or turning in bed
- Abstention with sensor displacement, device removal or missing samples
Surgical Environment Error Prevention
- Tracking and duplicate counts after smoke, occlusion or departure from the camera field
- A retained-item mismatch flagged before closure in synthetic sequences
- Single-frame latency of at most 33 ms on fixed hardware
Crisis prevention
Suicide Prevention Research for People with Mental Illness
- 100% traceability of fixed test decisions to source documents and spans
- Leakage from records created after the event
- Sensitivity to missing records and shortcuts based on age or diagnosis labels
Suicide Prevention Research for Bereaved People and Attempt Survivors
- Transition to the predefined pathway after an out-of-scope request
- Withholding or switching a response when records are incomplete or conflicting
- Separate denominators for follow-up scenarios involving bereaved people and attempt survivors
Social Isolation and Suicide Prevention Research
- Distinguish vacations, battery depletion and device non-wear from observed changes
- Check for storage of raw location or text fields
- Reject inferences drawn from unverified linkage of regional and individual data
Student Suicide Prevention Research
- Transition tests based on structured synthetic crisis markers
- Abstention with ambiguous markers and missing context
- Check for unrestricted responses after a handoff